anthropic research: WHICH JOBS ARE AT RISK OF AI?

Anthropic just released new research on which jobs AI is actually replacing and which ones it could replace in theory. They introduced this metric called "observed exposure." A new measure of occupational exposure. Our new measure, Observed Exposure, is meant to quantify: of those tasks that LLMs could theoretically speed up, which are actually seeing automated usage in professional settings. Theoretical capability encompasses a much broader range of tasks. By tracking how that gap narrows, observed exposure provides insight into economic changes as they emerge. In the chart you can see the blue area showing what AI could theoretically handle and the red area showing what it's handling today. For example, in computer and math jobs, AI could theoretically handle about 94% of tasks, but Claude is currently only covering about 33% of those in practice. Similar story for office and admin roles, with about 90% of tasks theoretically automatable, but actual usage today is just a fraction of that. So it's pretty clear that AI adoption is still very much so in its early days. And if you look at the top 10 most exposed occupations based off of this observed exposure metric, the number one is #1 computer programmers (74.5%)... of everything we're seeing around AI coding right now. #2 customer service reps (70.1%), data entry workers 67%, #3 data entry keyers (67.1%), medical record specialists 66%, #4 medical record specialists (66.7%), and you can read through the rest of the list. But you even have financial and investment analyst, #7 financial & investment analysts (57.2%), up there at 57%. One of the pretty clear themes is that all of these are knowledge work jobs. They're relatively well-paid and they often require higher education. In fact, the most AI exposed group in the study is generally more educated and also 16% more likely to be female. Which means that the workers most exposed to AI today are actually the very people who went through and pursued white collar careers, partly to avoid the kind of automation that hit manufacturing in the trades decades ago. It is important to keep in mind a few caveats here. This study combines real Claude usage data with estimates of what AI could theoretically do. So it's not necessarily a perfect prediction of job loss. On top of this, Anthropic says that despite the exposure numbers, we're not seeing mass unemployment yet. Their researchers actually compared workers in highly AI exposed jobs and workers in AI resistant jobs, and the difference in unemployment rate since the release of ChatGPT is basically statistically zero. So the feared AI job apocalypse that this report lays out hasn't shown up in the data, at least not yet.